Fraud Detection Using Segmented Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection techniques in online scenarios often result in false positives and false negatives, failing to accurately identify fraudulent activities.
Innovation Solution
A method utilizing supervised and unsupervised machine learning models to generate fraud scores and anomaly scores based on user verification attributes, allowing for the identification and differentiation of potential false positives and false negatives by segregating access requests into clusters and analyzing their similarity to historical requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fraud determination technique is used to identify fraud in online scenarios, then fraud detection capability is improved, but false positives and false negatives occur
Solution Approach 1:
The patent segments the fraud detection process into multiple independent components: supervised learning models for pattern recognition, unsupervised anomaly detectors for outlier identification, and ensemble methods for final decision-making. Each component operates on different aspects of the data and their results are combined to reduce false positives and false negatives, thereby improving both reliability and measurement precision simultaneously.
2Measurement precision
If multiple machine learning models are used to reduce false positives and false negatives, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The system divides the complex fraud detection task into separate modular components: supervised classifiers for known fraud patterns, unsupervised anomaly detectors for unknown patterns, and ensemble aggregation mechanisms. Each module is independently trained and evaluated, allowing for targeted optimization and easier maintenance while achieving high overall precision through their coordinated operation.
Data Source
AI summary
A method is disclosed. The method includes obtaining an access request associated with a user for a software application; obtaining a plurality of verification attributes associated with the user; generating a fraud score for the access request by feeding a supervised machine learning (ML) classifier with a feature vector for the user that is based on the plurality of verification attributes; selecting a first unsupervised ML anomaly detector of a plurality of unsupervised ML anomaly detectors based on the fraud score; generating an anomaly score for the access request by feeding the first unsupervised ML anomaly detector with an augmented feature vector for the user that is based on the plurality of verification attributes and the fraud score; and processing the access request based on the anomaly score.